Male- and female-specific reproductive risk factors across the lifespan for dementia or cognitive decline: a systematic review and meta-analysis | BMC Medicine

Search strategy and selection criteria

We followed the updated Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA 2020) statement guidance (9). Systematic searches were conducted in PubMed, EMBASE, and Cochrane Library, primarily through January 1, 2023, for studies focused on associations of female- and male-specific factors with dementia or cognitive decline. As there is already an extensive literature on studies of hormone replacement therapy (HRT) in women, this article will not address this topic (ten). Leverage previous research (11) and with the advice of experienced gynecologists and andrologists, we identified female-specific reproductive risk factors using the following search strategy: periods, placental bed disorders, menstrual cycle, polycystic ovary syndrome , reproductive period, menopause, climacteric symptoms, reproductive history, estradiol, sex. hormone-binding globulin, pregnancy, gravidity, parity, breast-feeding, intrauterine growth restriction, premature birth, stillbirth, induced abortion, gestational diabetes, ectopic pregnancy, hyperemesis gravidarum, gestational hypertension, ovarian hyperstimulation, in vitro fertilization, oophorectomy, hysterectomy, subfertility, cesarean section, natural labor, dystocia, fetal death, multiple pregnancies, postpartum hemorrhage, amniotic fluid embolism, puerperal infection, postpartum depression, hyperprolactinemia, amenorrhea, premature ovarian failure, hypothalamus-pituitary axis -ovary, test-tube baby, premature rupture of the membrane, intrahepatic cholestasis of pregnancy. The following strategy was used for male-specific reproductive risk factors: erectile dysfunction, testosterone, dihydrotestosterone, androstanediol glucuronide, androgen, androgen deprivation therapy, orchiectomy, Y haplotype cryptorchidism, prostatic hyperplasia, cryptorchidism, varicocele . The literature search strategy included the following outcome terms: Alzheimer’s, dementia, cognitive, and cognition. The primary outcome was dementia or cognitive decline. When multiple types of outcomes were reported for a factor, outcomes were categorized into subgroups of cognitive decline, dementia, and AD.

The inclusion criteria were as follows: (1) Studies had to be cohort studies. However, for female reproductive factors (menstrual factors, parity and gynecological operations), in addition to cohort studies, case-control studies were also included given the expectation of modest recall bias; (2) Outcome measurement had to be described in detail, with cognitive decline assessed using standard, large-scale cognitive tests, and dementia or AD diagnosed using objective, globalized diagnostic criteria such as Diagnostic and Statistical Manual of Mental Disorders criteria, International Classification of Diseases codes, National Institute of Neurological and Communicative Disorders and Stroke and Alzheimer’s Disease and Related Disorders Association criteria; and (3) For dose-response analysis, the grouped related factors had to be separated into at least three levels, with specific or calculable person-years and numbers of cases within each level.

Two researchers independently reviewed the included studies. If there was a discrepancy, the third author was asked to determine whether to include or exclude the study. When two or more studies came from the same database, the study with the largest sample size and/or the most detailed information was retained. Additionally, we combed through the bibliographies of qualified studies to avoid overlooking any potentially relevant studies.

Data Extraction

Predefined models were used to extract data from each article, including first author, year of publication, study design, population resources, cognitive status at baseline, mean age, gender composition, duration of follow-up, attrition rate, total sample size for analysis. incident case, outcome type, outcome measure, exposed factor type, exposed factor measure, adjusted confounders, and risk estimates. If required data were not reported in the publication, we contacted the authors to obtain them. Two authors with extensive experience extracted the data and any disagreements were resolved with the assistance of a third reviewer.

Assessment of study quality

The Newcastle-Ottawa Scale (NOS) was used to assess potential bias. To more accurately measure potential bias in the studies, a modified version of the NOS was used (12) (Additional file 1 Annex A and Annex B). The NOS can fully evaluate a single study, integrating various criteria such as representativeness, comparability, objectivity and reliability (Supplementary file 1: Appendix C).

statistical analyzes

Importantly, risk estimates and 95% confidence intervals (CIs) for a series of risk factors were acquired for further analysis. When odds ratios (ORs) were provided in some articles instead of relative risks (RRs) or hazard ratios (HRs), we used the following algorithm to convert ORs to RRs (13):

P.₀ represents the incidence of dementia or cognitive decline in the unexposed group. If P.₀ cannot be calculated, the overall incidence rate of the entire sample can be used instead.

To begin, we specified exposure definitions to facilitate comparison of pooled data across studies: (a) menarche was defined as age at menarche ≤ 13 years; (b) late first period was ≥ 16 years; (c) early menopause was ≤45 years; (d) late menopause was ≥54 years; (e) the short reproductive period was ≤ 34 years; (f) long reproductive period was ≥ 38 years; (g) early pregnancy was ≤ 20 years; and (h) late pregnancy was ≥ 30 years. The fixed model combined risk estimates from the same category within a study, while the random model pooled estimates across studies (13). Heterogeneity was assessed by the Q test and quantified by the I² metric (14). For factors with ≥10 studies, subgroup analysis and meta-regression were performed.

Dose-response analysis for eligible components was performed using inverse variance weighted least squares regression with clustered robust error variances (REMR model) (15). For studies that did not use the lowest category as the reference group, we reassigned the reference group and recalculated effect sizes using the Orsini method (12). When a range was provided, the midpoint represented the average exposure level. For open categories, the exposure level was set at the limit plus/minus the interval length of adjacent groups (16). Figures and analyzes were performed using GraphPad Prism 9.0 and Stata version 12.0.

Assessment of certainty of evidence

GRADE Notes

Five domains were used to assess the credibility of the meta-analysis: “risk of bias (17),”,”inconsistency (18),”,”imprecision (19),, “the indirect nature (20)” and “publication bias (21).” The certainty level of each domain was categorized as “0 (probably high), − 1 (probably moderate), or − 2 (probably low). The GRADE (Grading of Recommendations Assessment, Development, and Evaluation) approach assessed the overall credibility of the meta-analysis (22) (GRADE website: https://community.cochrane.org/help/tools-and-software/gradepro-gdt) (Additional file 1: Appendix D).

Systematic review index

Unlike traditional studies which neglected research not suitable for meta-analysis, we introduced a new parameter called “S index”. The S index was calculated using the following formula:

NOT refers to the total number of studies included in the systematic review. To better describe the results, we calculated both the “S index_For» and “S index_against.” S index_For reflected the number of high-quality studies consistent with the meta-analysis results, while the S index_against reflected those who disagreed. When calculating the S index_For, P. was 0 if the study results were not consistent with the meta-analysis, and 1 if they were consistent. Calculating the S index_against was reversed. Based on this, we introduced the concept of “S-index divergence”, calculated as the S-index_For minus the index S_against. A higher discrepancy score indicates greater support for the meta-analysis result (23).